To analyse the resolution and accuracy of cross-borehole ground penetrating radar (XBGPR) methods in monitoring the variation due to fluid infiltration in the heterogeneous vadose zone, we built a hypothetical two-dimensional model by using the Sandia-Tech Vadose Zone (STVZ) model for unsaturated-flow modelling. The spatial variation in water content provided by the flow modelling was converted to a dielectric constant and electrical conductivity. We then used these parameters in finite-difference time-domain (FDTD) electromagnetic (EM) forward modelling to simulate the response of cross borehole radar surveys with antenna configurations identical to those of the STVZ site. We inverted the synthetic GPR data by using a damped least-squares method with straight ray-path assumptions. The results from the resolution analysis suggest that attenuation within a clay layer is overestimated near the boreholes but underestimated between the boreholes in the inverted images synthesized by FDTD modelling. Compared to the attenuation images, inverted water-content images are more representative of the input model. The inverted attenuation also shows that artefacts such as 'bull's eye' structures appear in the attenuation images. The reason for the poorer resolution and artefacts in the attenuation images concerns the straight-ray assumptions that we made in the inversion to approximate propagation of the EM waves with additional energy loss when they crossed the velocity boundaries. In addition, the clay layer serves as a waveguide when both the transmitter and receiver located in the clay layer and our current inversion algorithm do not account for the waveguide physics. Thus, it is necessary to properly incorporate physics into the inversion algorithms in order to correctly invert for the attenuation.